...

For instance, in Figure 5 it looks like a threshold of 2 can get us what we want. Since the x axis is on the log2 scale, the corresponding value for "Lowest average coverage" is 2^2=4 (Figure 6). After we set the filter that way and rerun GSA (Figure 7), the shrinkage plots takes the required form (Figure 78).

Note that it is possible to achieve a similar effect by increasing a threshold of "Lowest maximal coverage", "Minimum coverage", or any similar filtering option (Figure 67). However, using "Average coverage" is the most straightforward: the shrinkage procedure uses log2(Average coverage) as an independent variable to fit the trend, so the x axis in the shrinkage plot is always log2(Average coverage) regardless of the filtering option chosen in Figure 67.

As suggested above, Lognormal with shrinkage is not intended to work for a dataset that contains a lot of zeros, a set of low expression features being one example. In that case, a count-based (Poisson or Negative Binomial) model is more appropriate. With that in mind, it appears to be a simple solution to enable Poisson, Negative Binomial, and Lognormal with shrinkage in Advanced Options and let the model selection happen automatically. However, there are a couple of technical difficulties with that approach. First, shrinkage can fail altogether because of the presence of low expression features (in that case, the user will get an informative error message in GSA). Second, there is an issue of reproducibility discussed above. However, for a large sample size shrinkage becomes all but disabled and reproducibility is not a concern. In that case, one can indeed cover all levels of expression by using Poisson, Negative Binomial, and Lognormal together.

...